Road user tracking using a Dempster-Shafer based classifying multiple-model PHD filter

Multi-object tracking requires appropriate motion models to predict the objects' states. In case of road user tracking, objects with different motion characteristics have to be concerned. Moreover, the motion characteristics and with that the appropriate motion model depends on the object's class. In this contribution a classifying multiple-model probability hypothesis density filter based on Dempster-Shafer theory is proposed. The object class is estimated based on features of the measurement as well as features of the estimated objects' states. Furthermore, the transition probabilities between the model modes are not static, but adapted with the estimated class probabilities of each track. It is shown, that a single multiple model filter is able to track multiple road users with different motion characteristics. Additionally, the integration of the Dempster-Shafer based classification in the filter framework improves the object class estimation significantly. Finally, an application of the filter on real world data of an intersection perception system is presented.